Characterization of Vehicle Behavior with Information Theory

TL;DR

This paper applies Information Theory and the Bandt-Pompe methodology to analyze vehicle velocity time series from diverse datasets, identifying traffic regimes like congestion and free flow through complexity-entropy analysis.

Contribution

It introduces a novel approach using Information Theory to characterize vehicle behavior and traffic regimes from velocity data across multiple real-world datasets.

Findings

Traffic velocities resemble correlated noise with power-law spectra.

Congestion correlates with near-random velocity patterns.

Free traffic flow shows more correlated velocity behaviors.

Abstract

This work proposes the use of Information Theory for the characterization of vehicles behavior through their velocities. Three public data sets were used: i.Mobile Century data set collected on Highway I-880, near Union City, California; ii.Borl\"ange GPS data set collected in the Swedish city of Borl\"ange; and iii.Beijing taxicabs data set collected in Beijing, China, where each vehicle speed is stored as a time series. The Bandt-Pompe methodology combined with the Complexity-Entropy plane were used to identify different regimes and behaviors. The global velocity is compatible with a correlated noise with f^{-k} Power Spectrum with k >= 0. With this we identify traffic behaviors as, for instance, random velocities (k aprox. 0) when there is congestion, and more correlated velocities (k aprox. 3) in the presence of free traffic flow.